Image Fusion In WMSN Based On Sparse Representation

Wireless Multimedia Sensor Network (WMSN) perceives the physical world through agroup of distributed multimedia sensor nodes. WMSN is applied widely as its advantages ofad hoc network, multi-path data transmission, high scalability and quick arrangement aboutnodes. However, the limited energy should not be ignored while transferring a large amountof images data; Meanwhile, the change of external environment leads to the acquisition imagecontaining a lot of background noise. It is necessary to use the appropriate image processingtechnology in WMSN.The infrared image has the strong ability of target recognition, it is not subject to externalenvironment conditions, but image signal to noise ratio is low, inadequate backgroundinformation; And visible light image resolution is higher, usually reflects the detailedinformation of the target scene, matches with the visual properties of humans, but its targetcharacteristics is not obvious under the condition of low contrast or illumination. In order toguarantee the reliability of the monitoring, we can collect infrared image and visible lightimage at the same time. And use the image fusion technology to fuse infrared and visible lightimage, which can get more accurate and comprehensive analysis of the target scene; Applyingthis technology in WMSN, not only can reduce the network transmission of data to a certainextent, also realize the image acquisition of all-weather. Traditional methods of image fusionis to deal with all pixels or coefficient, the amount of fused data is larger. And they did notmake full use of the image structure characteristics and inherent sparse. It is difficult to get thehigh quality of fused WMSN image.In recent years, compressed sensing theory is for the attention of many scholars, especiallyon sparse representation theory of super sparse complete dictionary is widely used in all kindsof image processing, such as fusion, recovery, denoising, etc. For the problem of limitedenergy in WMSN, this paper proposes a WMSN image fusion method based on DWT sparsematrix. The method uses the inner sparse of the WMSN image under the framework ofcompressed sensing technology. In the process of fusion, it only needs to fuse a small numberof measurements by improving the fusion rules, then reconstructs the image. But whenexternal environment is getting worsen, WMSN image will contain lots of gaussian whitenoise, the method will be difficult to get the high quality of fused image from the low PSNRsource image.In order to solve the above problems, this paper proposes the fusion method based oncomplete dictionary sparse. K-SVD training algorithm is more effective for image sparserepresentation. In this method, we combines with K-SVD training algorithm on the basis of the original DCT redundant dictionary, and designed the corresponding fusion rules for a fewsparse coefficient, then get the fused sparse coefficients. The method can decrease the amountof data transmission and get rid of much noise, moreover, it fully retains useful information ofthe WMSN image.